this method was not used, not documented and not tested.

Rename `ActiveRecord::Fixtures` class to `ActiveRecord::FixtureSet`.  Instances of this class normally hold a collection of fixtures (records) loaded either from a single YAML file, or from a file and a folder with the same name.  This change make the class name singular and makes the class easier to distinguish from the modules like `ActiveRecord::TestFixtures`, which operates on multiple fixture sets, or `DelegatingFixtures`, `::Fixtures`, etc., and from the class `ActiveRecord::Fixture`, which corresponds to a single fixture.

If you create a new record via a collection association proxy that has
not loaded its target, and which selects additional attributes through
the association, then when the proxy loads its target, it will
inadvertently trigger an ActiveModel::MissingAttributeError during
attribute writing when CollectionAssociation#merge_target_lists attempts
to do its thing, since the newly loaded records will possess attributes
the created record does not.

This error also raises a bogus/confusing deprecation warning when
accessing the association in Rails 3.2.x, so cherry-pick would be
appreciated!

This reverts commit abf8de85.
We should take a deeper look to those cases flat_map doesn't do deep
flattening.

irb(main):002:0> [[[1,3], [1,2]]].map{|i| i}.flatten
=> [1, 3, 1, 2]
irb(main):003:0> [[[1,3], [1,2]]].flat_map{|i| i}
=> [[1, 3], [1, 2]]

- Changed the comments documentation for the update_column(s) methods to add a little bit of clarity

Patches `CollectionAssociation#count` to return 0 without querying
if the parent record is new. Consider the following code:

    class Account
      has_many :dossiers
    end

    class Dossier
      belongs_to :account
    end

    a = Account.new
    a.dossiers.build

    # before patch
    a.dossiers.count
    # SELECT COUNT(*) FROM "dossiers" WHERE "dossiers"."account_id" IS NULL
    # => 0

    # after
    a.dosiers.count # fires without sql query
    # => 0

Fixes #1856.

The counter column name in the intermediate model need to be access
via the through reflection.

reflection init as 'if' stantment.
column is always the same expression and depends from the changing of attributes

When inserting new records, only the fields which have been changed
from the defaults will actually be included in the INSERT statement.
The other fields will be populated by the database.

This is more efficient, and also means that it will be safe to
remove database columns without getting subsequent errors in running
app processes (so long as the code in those processes doesn't
contain any references to the removed column).

Fix destructive side effects from marshaling an association caused by 65843e1a